Browse Prior Art Database

Simultaneous Morphological Transforms of Interdependent Binary Images

IP.com Disclosure Number: IPCOM000102492D
Original Publication Date: 1990-Nov-01
Included in the Prior Art Database: 2005-Mar-17
Document File: 4 page(s) / 195K

Publishing Venue

IBM

Related People

Kimmel, MJ: AUTHOR

Abstract

Disclosed is a method for efficient, functional simulation of networks of morphological or neighborhood image transforms. Existing methods such as truth table lookup (1) cannot exploit bit parallelism. Bit parallelism is possible in restricted cases such as multiple independent images, or with a single transform on a single image (2). In contrast, this approach can utilize bit parallelism in all situations including iteration and feedback. The solution is useful for simulating complex algorithms on configurable parallel pipeline image processing systems (3,4), and for directly computing iterative transforms such as required for skeletonization (5) or thinning (6).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 35% of the total text.

Simultaneous Morphological Transforms of Interdependent Binary Images

       Disclosed is a method for efficient, functional
simulation of networks of morphological or neighborhood image
transforms.  Existing methods such as truth table lookup (1) cannot
exploit bit parallelism. Bit parallelism is possible in restricted
cases such as multiple independent images, or with a single transform
on a single image (2).  In contrast, this approach can utilize bit
parallelism in all situations including iteration and feedback.  The
solution is useful for simulating complex algorithms on configurable
parallel pipeline image processing systems (3,4), and for directly
computing iterative transforms such as required for skeletonization
(5) or thinning (6).

      Networks or sequences of image primitives are subsequently
referred to as Operation Flow Graphs (OFGs). The neighborhood
transform primitive is performed by a Processing Element (PE).  The
MITE is a typical configurable parallel pipeline image processing
system to which this simulation technique can apply.  See (3) for
more details on the MITE and for more succinct definitions.

      This technique preserves the time equivalence between different
images exactly as the machine hardware of MITE. The multiplicity of
input bits for bit parallelism come from among equal time pixels
within the neighborhood windows of PEs which are performing the same
transform. For example, the northwest pixels from many images at a
specific time are in the same word.  The simulation program advances
all PE outputs (one bit each) to the input(s) of the next PE
synchronously.  True implementation delays are preserved.

      An iterative portion of an OFG, shown in Figure 1, demonstrates
the technique.  Each of the N PEs (ITER_1 .... ITER_N) perform the
same transform (TYPE=A).  Delay within a PE holds enough bits of the
image to form the desired window.  The delay bits can be considered
as part of the internal state of the PE.  Each PE has an internal
state of M bits.  M will be the delay which is equal to twice the
image width in pixels plus three, i.e., 2W+3, plus the number of
feedback bits internal to the PE.  In the usual implementation of a
PE with a three- by-three window there is no feedback.  In the MITE
implementation there is just one, shown as FB in Figure 5 of the MITE
paper [3].  The M internal state bits for N PEs are stored in a block
of memory at least as large as M words, each word containing at least
N bits. One bit from each of the N image streams flowing through the
N PEs in the iteration comprise the word.  Figure 2 shows a block of
memory larger than M words with some of the words labelled by their
neighborhood position relative to the same reference pixel - in
identical network time rather than equivalent spatial location in the
image.  The word labelled "center" is identical bit for bit to the
"center" state bits of the N PEs in Figure 1 for each system clock...